This invention relates generally to avalanche photodiodes (APDs) and in particular to monolithic planar arrays of APDs and methods for forming such arrays.
Devices used in radiation imaging for medical purposes or the like must be able to detect low levels of incident optical photons or ionizing particles to minimize patient exposure to radiation. In such imaging devices it is often advantageous to employ radiation detection devices having internal gain; avalanche photodiodes (APDs) are commonly used in such devices to provide the desired detection sensitivity. An APD is a semiconductor device that is biased near the breakdown region such that charge generated as a result of the absorption of an incident photon is amplified in the APD itself as a result of a cascading effect as charge is accelerated by the high bias potential applied across the p-n junction of the device. In such imaging devices, it is desirable that the APD exhibit low noise and high gain. Certain devices, such as medical imagers (e.g., using gamma radiation), also require relatively large arrays (e.g., about 5 cm.sup.2 or larger) of high quality, low noise APDs.
Currently, the two types of APD designs in common use are the "deep diffused" structure and the "reach through" structure. Deep diffused APDs exhibit a wide avalanche region and operate at a relatively small electric field, resulting in a small value of the device parameter "k" (which determines excess noise), large gain, and stable device operation.
Deep diffused APDs, however, are typically not readily manufactured in large arrays as each device must be formed to have a precise bevel at the edge of the device. The bevel is required to reduce the peak surface field (i.e., the electric field across the p-n junction in the area where the p-n junction intersects the surface of the structure) of the APD well below the peak bulk electric field (i.e., the electric field across the p-n junction in the body of the device where the p-n junction is disposed substantially parallel to the surfaces of the device to which the bias is applied) so that the APD breaks down in the bulk instead of at the surface. For example, the peak surface field should have a value about 70% or less than the value of the bulk field to ensure the APD breaks down in the bulk (breakdown outside of the bulk results in significantly more device noise).
Bevel formation requires mechanical operations which make the fabrication process non-standard in that the bevel for each device must be individually formed. The non-standard methods required for bevel formation also results in reduced yield and non-uniform reliability of the devices formed, making the fabrication of large area arrays of this type of device expensive and difficult. Furthermore, the beveled edge procedure is difficult to extend to monolithic arrays because of the lack of a suitable isolation technique between adjacent devices on a chip.
Deep diffused APDs are typically fabricated by diffusing a p type dopant, such as gallium, into a wafer of n type material from both sides of the wafer, resulting in two parallel p-n junctions in the wafer. The p doped material on one side of the wafer is completely removed so that only one p-n junction remains in the wafer. The remaining p and n regions are thinned to appropriate dimensions and p+ and n+ diffusions are made to respective sides of the wafer to form a contact pad to the p and n doped sections, respectively. Discrete devices are then diced from the wafer, and the edges are beveled as described above to obtain the desired surface field characteristics for good reverse blocking capability. The mechanical cutting and finishing (such as by etching) of the bevels renders an array of such devices mechanically fragile. As noted above, such a process is time consuming and requires a high degree of precision to form the appropriate bevels, and such a structure requires a careful passivation to minimize the injection of charge from the beveled portion of the device.
Variations of this technique, in which shallow grooves are cut that extend close to, but not into, the depletion area of the device, have also been tried. Isolation in such a device is a function of lateral inter-pixel resistance as compared to the input resistance of the readout circuit; because the depletion layer changes depending upon applied voltage, isolation in this type of device is obtained only for a very narrow range of applied bias voltage. Further, even in this narrow range of voltage, a small input resistance is needed in the readout circuit to provide adequate isolation. The Johnson noise from this resistance, because of its small value, typically introduces unacceptable noise levels in the preamplifier and is unacceptable for medical imaging devices.
The reach through APD structure generally does not require bevel formation. The reach through type of APD typically has a shallow p-n junction that results in lower gain, a larger value of k (resulting in high noise devices), and greater temperature drift than deep diffused devices. Further, the active area of reach through devices is small as compared to deep diffused devices. Array fabrication can be accomplished, although the process is time consuming and expensive as many steps are required to fabricate the array, and the resulting APDs in the array suffer the drawbacks noted above. Arrays in reach-through technology are also limited to a small active area.
For most imager devices, it is thus desirable to have an APD array that is readily fabricated and that contains high quality individual APD pixels, that is APDs that exhibit low noise and high gain. It is also desirable that the array be structurally strong.
It is an object of this invention to provide an APD array that is structurally strong, readily fabricated, and that comprises APDs that exhibit low noise and high gain.
It is a further object of this invention to provide a method of fabricating a planar APD structure comprising high quality APDs.